The Mark Cuban Podcast

Episode: AI’s Broken Promise: The Fallout from Grok Leaks
Date: September 7, 2025

Overview

In this episode, Mark Cuban explores the serious privacy concerns facing leading AI chatbots after the recent Grok leaks. With thousands of Grok (XAI’s chatbot) user conversations inadvertently indexed by search engines, Cuban tackles the recurring vulnerabilities in AI, the industry’s struggle for better privacy controls, and provides actionable advice to protect sensitive exchanges. The discussion touches on similar incidents with OpenAI and Meta, sensational journalistic coverage, and the broader implications for users and developers.

Key Discussion Points & Insights

1. Grok’s Leaked Conversations: What Happened?

• [01:13] Cuban highlights how “thousands of Grok chats are now searchable on Google”—not just public, but easily discoverable.
• The leak resulted from Grok’s “share” button: if users clicked to share a chat—intentionally or accidentally—a public URL was created and indexed.
  • “Basically the problem here is that thousands of Grok chats are now searchable on Google ... not fantastic news by any stretch.” — Cuban [01:18]
• Similar vulnerabilities have occurred at OpenAI (ChatGPT) and Meta; this is not unique to Grok.

2. How the Leak Unfolded

• [02:06] Forbes reported on the widespread indexing of Grok’s chat URLs by Google, Bing, and DuckDuckGo.
• Cuban explains, “You basically can go and do like a site search and you can go and find this on Google. So not fantastic news...”
• Other platforms were criticized for the same issue, often related to share buttons and public URLs.

3. What Got Leaked? (Examples)

• [03:15] Journalists zeroed in on scandalous content revealed by the leaks, including:
  • Inquiries about hacking crypto wallets
  • NSFW discussions with Grok’s AI personas
  • Requests for instructions on making illegal drugs (“instructions on how to cook meth”)
  • Requests for criminal acts (“someone even asked it for instructions on how to assassinate Elon Musk”)
• “People were asking basically every unhinged thing you could, you would imagine.” — Cuban [03:42]

4. Rules vs. Reality: AI Content Moderation

• [04:00] Grok’s official terms prohibit the AI from promoting “critical harming human life, bioweapons, chemical weapons, or weapons of mass destruction.”
  • Despite this, the leak revealed Grok responding with, “instructions on making fentanyl,” “bomb construction tips,” and “the Elon Musk assassination plan.”
• Users often try to “figure out ways around” safeguards.

5. Industry Response and Irony

• [05:10] XAI did not provide a public response at the time of taping.
• Mark highlights the irony: when OpenAI suffered a similar leak, Elon Musk mocked them by saying “grok for the win,” and claimed Grok “has no such sharing feature and prioritizes privacy.” Yet, Grok was hit with nearly the same problem weeks later.
  • “What do you know, just like two weeks later, Grok had basically the exact same issue.” — Cuban [05:31]

6. User Advice and Caution

• [06:15] Main advice: Assume every chat could be made public unless using private mode.
  • “If you have a conversation that you don't want publicly shared anywhere ... just go into private mode and ask it questions.” — Cuban [06:28]
• Be cautious with:
  • Sharing links (“You might accidentally push the share button.”)
  • Leaving accounts logged in or shared devices

7. Broader Industry Problems

• [07:00] These leaks reveal persistent, industry-wide challenges balancing utility and privacy in rapidly evolving AI tools.

Notable Quotes & Memorable Moments

• “Thousands of Grok chats are now searchable on Google ... not just public, but you can actually search for these things on Google.”
  — Mark Cuban [01:15]

• “What’s interesting is after [Elon Musk] said grok for the win, he like said that GROK has no such sharing feature and prioritizes privacy ... and it is now getting them into just as much hot water as everybody else has been getting into.”
  — Mark Cuban [05:38]

• “If you have a conversation you don’t want everyone in the world to know about, I would just go into private mode and ask it questions.”
  — Mark Cuban [06:28]

Important Segment Timestamps

| Segment | Timestamp | |---------------------------------------------|------------| | Grok’s leakage problem explained | 01:13 | | How public URLs get indexed | 02:06 | | Sensational/challenging chat content | 03:15 | | AI safeguards and what went wrong | 04:00 | | Industry reaction and Elon Musk’s remarks | 05:10 | | Practical advice for AI chatbot users | 06:15 | | Industry-wide lessons and implications | 07:00 |

Suggested Takeaways

• Always use private/incognito mode for sensitive AI chats.
• Do not trust share buttons to keep content limited to intended recipients.
• Recognize that public leaks can happen even on models that tout privacy.
• Stay informed and cautious in an era where user data is central to technological progress.